MTVQA

MTVQA是首个针对多语言文本中心视觉问答（TEC-VQA）场景提供高质量人类专家标注的基准数据集。该数据集包含9种不同语言的28,607个问题-答案对，涵盖了从简单内容提取到文本相关推理的多样化问题类型。数据集中的图像来自真实世界，经过精心筛选和标注，确保视觉文本对齐。MTVQA旨在评估和提升多语言环境下AI模型在文本丰富场景中的理解和回答能力，特别关注低资源语言的处理，为全球社区提供了一个独特的多语言VQA资源。

MTVQA is the first high-quality human-annotated benchmark dataset for the multilingual Text-Centered Visual Question Answering (TEC-VQA) scenario. This dataset contains 28,607 question-answer pairs across 9 distinct languages, covering diverse question types ranging from simple content extraction to text-related reasoning. The images in the dataset are sourced from real-world scenarios, carefully filtered and annotated to ensure visual-text alignment. MTVQA aims to evaluate and enhance the ability of AI models to understand and answer questions in text-rich multilingual environments, with a special focus on low-resource language processing, providing a unique multilingual VQA resource for the global community.